ondrej-merkun/skill-audit
Scan AI agent skills (Claude Code, Cursor, Copilot, Codex, Gemini, Windsurf, Cline) for prompt injection and malicious code.
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Activelast commit May 12, 2026
- License
- None
Pinned Snippet
uses: ondrej-merkun/skill-audit@836d3e00ed562834380e5d799a771de2aa57be0d # v0.2.1tags can be moved; commit SHAs can't. why a SHA?
Inputs
| name | description | required | default |
|---|---|---|---|
| version | @ondrej-merkun/skill-audit npm package version to run (e.g. "0.2.1" or "latest") | no | latest |
| path | Working directory to scan (default: repo root) | no | . |
| args | Additional arguments passed verbatim to `skill-audit scan` | no | "" |
| fail-on | Exit 1 when any result meets this verdict band or worse. One of: review, fail | no | fail |
| agent | Restrict discovery to one agent id (claude-code, cursor, copilot, cross-agent). Leave blank for all agents. | no | "" |
| upload-results | Upload JSON results as a workflow artifact | no | true |
| results-file | Path where the JSON results file is written | no | skill-audit-results.json |
Outputs
| name | description |
|---|---|
| results-file | Path to the JSON results file |
| verdict | Worst verdict across all scanned skills (PASS, REVIEW, FAIL) |